The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses npx to fetch and execute the @agent-native/core package and hosted MCP connectors from the vendor's repository to establish the visual editing environment.
[COMMAND_EXECUTION]: Instructions direct the agent to execute shell commands to start a local bridge daemon on port 7331. This bridge manages the connection between the local development server and the remote design editor.
[DATA_EXFILTRATION]: The skill reads local project files and sends application metadata (routes, connection IDs) to the vendor's service at https://design.agent-native.com. It includes built-in safeguards that prevent access to sensitive files such as .env, .git, and private keys.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests local application content and source code to render its editing interface.
Ingestion points: Local source code files and running application routes/DOM state in SKILL.md.
Boundary markers: Defines a "hard write boundary" for reprompting actions to limit the scope of AI-driven changes.
Capability inventory: Possesses file system write capabilities (write-local-file) and the ability to manage local network bridges.
Sanitization: Implements a blocklist for sensitive file paths and requires human-in-the-loop consent for write operations.

visual-edit